Simulated data were used to develop theoretical indices with various intracorrelation and intercorrelation structures of the components and outcomes. Moreover, dietary intake data of 668 elderly people from the MEDIS (Mediterranean Islands) study were also used to develop a diet index and to test it toward obesity status (outcome).
On the basis of 1,000 simulations, we observed that the diagnostic accuracy of an index increases as the number of components increases, only when the components are not intercorrelated or have low intercorrelation. Moreover, the diagnostic accuracy of an index developed with all components associated with an outcome is higher compared with an index developed by using only some components related to the outcome. Finally, the predictive ability of an isolated component is lower than that of an index developed by using non-intercorrelated or low-intercorrelated components. Real data confirmed the aforementioned findings.
Low-intercorrelated or non-intercorrelated components, strongly associated with a particular outcome, should be used in order to obtain an accurate composite index.